A feature extraction-based cutting force fluctuation model construction method

Abstract

The application provides a feature extraction-based cutting force fluctuation model construction method. The application firstly performs data preprocessing on a monitoring signal, then performs signal space-time discretization, subdivides the monitoring signal along the time direction, defines cutting force fluctuation features, selects main features therefrom, extracts time domain and frequency domain features of data and selects auxiliary features therefrom, constructs a cutting force fluctuation model based on a Pearson correlation coefficient method, performs feature fusion on the main features and the auxiliary features, finally adopts a time sequence matching and coordinate matching method to associate cutting force fluctuation information with working condition information, constructs an association model of the cutting force fluctuation information and the working condition information, so as to realize accurate monitoring of cutting force fluctuation in a machining process, provide inspiration for application and research in a similar mass cutting force data field, improve data visualization in the application of monitoring a numerical control machining process, and ensure workpiece quality.

Description

Technical Field

[0001] This invention belongs to the field of CNC machining technology, specifically relating to a method for constructing a cutting force fluctuation model based on feature extraction. Background Technology

[0002] The CNC machining process is a complex metal cutting process, and the generation of cutting force is a crucial physical phenomenon. Cutting force primarily originates from the elastic and plastic deformation of the workpiece material, as well as friction between the cutting tool and the workpiece surface. With the development of automatic control technology, cutting force has become a vital parameter for controlling the machining process. Changes in cutting force not only affect the stability of the machining process but also directly determine the quality of the machined parts.

[0003] By monitoring changes in cutting forces, information such as workpiece chatter, cutting parameters, machine tool malfunctions, and tool wear can be obtained in real time, facilitating timely control of the cutting process. However, with the continuous improvement of CNC machine tool performance in recent years, sensor technology has also developed rapidly, enabling data acquisition at higher speeds and frequencies during machining. Faced with this massive and redundant cutting force data, directly analyzing the trend of cutting force changes and constructing a monitoring model of cutting force values ​​along the time dimension using traditional methods would sacrifice a large number of minute cutting force fluctuation characteristics in the data, making it difficult for the monitoring model to filter and extract the complex patterns and trends described in the massive data. Although sensor technology has improved, the value of the data received by the sensors is not fully utilized, ultimately failing to provide timely, effective, and more accurate guidance for the machining process.

[0004] To address the above problems, this invention proposes a method for constructing a cutting force fluctuation model based on feature extraction. Summary of the Invention

[0005] Because conventional data processing methods directly analyze the trend of cutting force changes and use this to construct a monitoring model of cutting force values ​​along the time dimension, they sacrifice the massive amount of minute cutting force fluctuation characteristics, resulting in low efficiency and difficulty in providing timely and effective guidance for the machining process. To solve the problem of processing massive cutting force data and extracting cutting force fluctuation characteristics, this invention proposes a cutting force fluctuation model construction method based on feature extraction.

[0006] To achieve the above objectives, the technical solution provided by this invention is:

[0007] A method for constructing a cutting force fluctuation model based on feature extraction, characterized by the following steps:

[0008] Step 1: Preprocess the raw cutting data;

[0009] Step 2: Subdivide the cutting force signal, tool path, and operating condition signal;

[0010] Step 3: Define the cutting force fluctuation characteristics;

[0011] For each short time domain T i The fluctuation characteristics of the cutting force of all micro-elements are expressed by combining "main features and auxiliary features", including the following sub-steps:

[0012] Step 3.1: Among all the fluctuation characteristics of micro-element cutting force, select several fluctuation characteristics with the best data dispersion, and select one of them as the main feature of micro-element cutting force fluctuation;

[0013] Step 3.2: Among all the fluctuation characteristics of micro-element cutting force, select the features that are highly correlated with the main features selected in Step 3.1. Then, select the features among these features whose correlation with the main features in Step 3.1 reaches the set requirements as auxiliary features of micro-element cutting force fluctuation, and define the feature parameters and expression formulas of the auxiliary features.

[0014] The auxiliary features include time-domain features and frequency-domain features;

[0015] Step 4: Use feature fusion methods to describe cutting force fluctuations and construct a cutting force fluctuation model;

[0016] Step 5: Use time-series matching and coordinate matching methods to establish a correlation model between cutting force fluctuation information and working condition information.

[0017] Further, step 1 includes the following sub-steps:

[0018] Step 1.1: Handle missing values ​​in the cutting data;

[0019] Step 1.2: Standardize the raw cutting data;

[0020] The conversion formula is:

[0021]

[0022] Where: x represents the original data; x′ represents the processed data; σ represents the standard deviation of the original data;

[0023] Step 1.3: Process the noise value of the cutting force signal.

[0024] Furthermore, step 2 includes the following sub-steps:

[0025] Step 2.1: Subdivide the cutting force signal according to the time-domain discretization concept;

[0026] Discretize the time domain of a processing procedure into N short time domains T with equal time intervals. i The discretized short-time domain set is as follows:

[0027] T(N)={T i |T1,T2,...,T N}

[0028] Where N is the number of short time-domain discrete points corresponding to the processing process, i = 1, 2, ..., N;

[0029] Step 2.2: Subdivide the tool trajectory condition signal according to the spatial discretization concept;

[0030] The machining process is discretized along the tool path signal into N equally timed cutting micro-elements S. i Then the set of cutting infinitesimal elements is:

[0031] S(N)={S i |S1,S2,...,S N}

[0032] N also represents the number of spatially discrete cutting micro-elements corresponding to the machining process, and is numerically equal to the number in the short time domain.

[0033] Furthermore, step 4 includes the following sub-steps:

[0034] Step 4.1: Construct the feature evaluation matrix based on the Pearson correlation coefficient method;

[0035] For all the fluctuation characteristics of the micro-element cutting force, first calculate the correlation coefficient of each characteristic with other characteristics, and then construct the feature evaluation matrix of all the micro-element cutting force fluctuation characteristics;

[0036] Step 4.2: Sort all auxiliary features from high to low according to their correlation coefficient with the main feature, and perform feature fusion on the main feature and auxiliary features to construct the cutting force fluctuation model.

[0037] Furthermore, step 5 includes the following sub-steps:

[0038] Step 5.1: Set the total data acquisition time, and discretize the cutting force data in the time domain according to the order of the machining process to obtain the data set of time-varying elements;

[0039] Among them, the time length, number and temporal sequence of the time micro-element are all matched with the short time domain discretization along the time dimension in step 2.1;

[0040] Step 5.2: Based on the read machine tool code, obtain the cutting tool path, and spatially discretize the cutting force condition along the tool path to obtain a set of condition volume elements;

[0041] The number of working condition elements is equal to the number of cutting micro-elements discretized along the tool path in step 2.2;

[0042] Step 5.3: By aligning the discrete cutting force time-domain data and the discrete working condition volumetric model in time, a correlation model between cutting force fluctuation information and working condition information is established.

[0043] Furthermore, the main feature in step 3.1 is variance.

[0044] Furthermore, the time-domain features in step 3.2 include:

[0045] Mean, root mean square value, peak value, kurtosis factor, peak factor, waveform factor, impulse factor, standard deviation, root square amplitude, peak-to-peak value, skewness, kurtosis, and margin factor;

[0046] The frequency domain features in step 3.2 include:

[0047] Centroid frequency, average frequency, root mean square frequency, and frequency variance.

[0048] The concept and principle of this invention:

[0049] like Figure 1 As shown, the monitoring signal is first preprocessed, including handling missing values, data standardization, and noise reduction, to ensure the quality of the original data. Then, the signal is spatiotemporally discretized, subdividing the monitoring signal along the time direction and timestamping each subdivided short-time domain. Next, cutting force fluctuation characteristics are defined, with variance selected as the primary feature. Time and frequency domain features are then extracted, and secondary features are selected. A cutting force fluctuation model is constructed based on the Pearson correlation coefficient method, and the primary and secondary features are fused. Finally, time-series matching and coordinate matching methods are used to associate the cutting force fluctuation information with the working condition information, constructing a correlation model between cutting force fluctuation information and working condition information to characterize the cutting force fluctuation characteristics.

[0050] The advantages of this invention are:

[0051] 1. This invention extracts cutting force fluctuation features through a main feature-auxiliary feature fusion method. Among all the fluctuation features of micro-element cutting force, several fluctuation features with the best data dispersion are selected, and one of these fluctuation features is selected as the main feature of the micro-element cutting force fluctuation. Based on this, a cutting force fluctuation model is constructed to achieve accurate monitoring of cutting force fluctuations during the machining process. This provides inspiration for applications and research in fields such as massive cutting force data, and has significant practical implications for monitoring CNC machining processes and ensuring workpiece quality.

[0052] 2. This method uses cutting force fluctuation as the standard for evaluating workpiece machining quality. By establishing a correlation model between cutting force fluctuation information and working condition information, it provides an accurate basis for assessing and making decisions on cutting force fluctuation. Based on the cutting force fluctuation during machining, it achieves smooth machining by timely controlling and adjusting cutting parameters, thereby improving machining efficiency and accuracy. Attached Figure Description

[0053] Figure 1 : Logical diagram of the method for constructing a cutting force fluctuation model based on feature extraction;

[0054] Figure 2 : Result of mean interpolation processing;

[0055] Figure 3 : Graph showing the results of standardizing the original cutting data;

[0056] Figure 4 Figure showing the data denoising results based on wavelet analysis;

[0057] Figure 5 Time-discrete plot of cutting force signal;

[0058] Figure 6 : Spatial discrete diagram of operating condition signals;

[0059] Figure 7 Thermographic diagram of correlation analysis of cutting force fluctuation characteristics of 18 micro-elements;

[0060] Figure 8 Visualization results of the correlation between the cutting force fluctuation characteristics of 18 micro-elements;

[0061] Figure 9 : A line graph showing the cutting force fluctuation characteristics after feature fusion;

[0062] Figure 10 Spatiotemporal correlation diagram of process signals and operating condition signals;

[0063] Figure 11 Schematic diagram of the correlation model between cutting force fluctuation information and working condition information. Detailed Implementation

[0064] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0065] Since cutting force is a very important process signal in the cutting process and can directly indicate the machining status, by monitoring the changes in cutting force during the cutting process, we can determine the wear or breakage of the tool, whether other cutting parameters are reasonable, etc. Therefore, monitoring cutting force can provide data basis for timely adjustment of cutting parameters to achieve smooth and efficient machining.

[0066] During machining, changes in cutting force manifest as fluctuations in the acquired force signal. A more stable machining process results in less noticeable signal fluctuations and better machining quality. Therefore, this method uses cutting force fluctuations as a standard for evaluating workpiece machining quality. Based on these fluctuations, cutting parameters are controlled and adjusted in a timely manner to achieve stable machining and improve machining efficiency and accuracy.

[0067] A method for constructing a cutting force fluctuation model based on feature extraction includes the following steps:

[0068] Step 1: Preprocess the raw cutting data;

[0069] Due to the complexity of the cutting process, the collected cutting data may suffer from problems such as inconsistent dimensions, incompleteness, high frequency, and significant noise interference. To improve data quality, preprocessing of the raw data is necessary.

[0070] The experimental data collected in this embodiment comes from the cutting force signal acquired by the Spike smart sensor force measuring tool holder during milling. The sensor acquisition frequency is 2500Hz. The cutting parameter combination is: number of milling cutter teeth N=3, spindle speed n=2000r / min, and feed rate f=300mm / min.

[0071] Includes the following sub-steps:

[0072] Step 1.1: Handle missing values ​​in the cutting data;

[0073] There are generally two methods for handling missing values: deleting individual samples containing missing values ​​and imputation. However, deleting missing values ​​results in a significant waste of information, so imputation is the preferred method, including mean imputation, multiple imputation, and maximum likelihood estimation.

[0074] This embodiment uses the mean interpolation method, and the sampling frequency of the data acquisition equipment used in the experiment is 2500Hz, that is, the sampling interval is 0.0004s. Figure 2 As shown, Figure 2 In Figure A, a portion of the data (the difference between the two data points within the box is too large) has missing segments. After processing the missing values ​​using mean interpolation, this portion of the data is as follows: Figure 2 As shown in Figure B.

[0075] Step 1.2: Standardize the raw cutting data;

[0076] When collecting processing data, the dimensions and orders of magnitude of the data may differ. When factors under different working conditions vary significantly, their weights in the model will also differ significantly. If the absolute value of the value is large, factors with little impact on processing may dominate, while some factors with low absolute values ​​may have a significant impact on the results, seriously confusing the relative importance of factors and making the model results unsatisfactory and unconvincing. Therefore, it is necessary to first analyze the collected data and standardize the data of various working conditions to eliminate the influence of dimensions and orders of magnitude.

[0077] Considering that the raw cutting force data collected by the sensor contains both positive and negative values, this embodiment uses the Z-score method to standardize the raw data, unifying the dimensions and orders of magnitude of the data while retaining all the information of the raw data.

[0078] The processed data have a mean of 0 and a standard deviation of 1.

[0079] The conversion formula is:

[0080]

[0081] Where: x represents the original data; x′ represents the processed data; σ represents the standard deviation of the original data;

[0082] The processing result is as follows Figure 3 As shown, Figure 3 Figure C in the image represents the original data, such as... Figure 3 Figure D in the figure shows the processed data, with the dimensions and order of magnitude of the data being unified;

[0083] Step 1.3: Process the noise value of the cutting force signal;

[0084] The cutting force signal during the cutting process is a polymorphic, unsteady signal, which includes random signals and interference signals.

[0085] Wavelet analysis has basis functions with a certain time and frequency resolution; therefore, this embodiment chooses wavelet analysis to denoise the cutting force signal. Based on the principle of wavelet analysis, the cutting force signal is decomposed into a series of superimposed wavelet functions at multiple scales, thereby achieving filtering and noise reduction.

[0086] The specific number of signal decomposition layers is determined based on the actual signal analysis results;

[0087] like Figure 4 As shown, the cutting force signal in this embodiment is decomposed into noise signal and reconstructed signal, which can achieve better processing results.

[0088] Step 2: To facilitate the study, the cutting force signal is subdivided.

[0089] The cutting force signal is generated along with the machining process, and the machining condition at each instant determines the cutting force at that instant. As the machining process continues, cutting force signals are continuously generated at different machining positions. Therefore, the cutting force signal has both time-varying characteristics F(t) and spatially variable characteristics F(p):

[0090] F(t)=[F(t1),…,F(t1),…F(t n )]

[0091] F(p)=[F(p1),…,F(p1),…F(p n )]

[0092] Since the cutting force signal is obtained by sampling at equal intervals, it is further subdivided based on the spatiotemporal characteristics of the cutting force signal and the ideas of time-domain discretization and spatial discretization.

[0093] Includes the following sub-steps:

[0094] Step 2.1: Subdivide the cutting force signal according to the idea of ​​time-domain discretization, and mark each short time domain with a timestamp to give the short time domain a time attribute;

[0095] like Figure 5 As shown, the time domain of a processing procedure is discretized into N short time domains T with equal time intervals. i Because of each short time domain T i The number of corresponding sampling points is related to the sampling frequency, therefore the short time domain T i Number of samples N i for:

[0096] N i =T i f

[0097] Where f is the sampling frequency;

[0098] The discretized short-time domain microdata set is as follows:

[0099] T(N)={T i |T1,T2,...,T N}

[0100] Where N is the number of short time-domain discrete points corresponding to the processing process, i = 1, 2, ..., N.

[0101] Step 2.2: Subdivide the tool path and working condition signals according to the spatial discretization concept;

[0102] like Figure 6As shown, the machining condition signal is discretized in space, that is, the material removed at each short time corresponding to the tool cutting trajectory is regarded as a micro-element. In this way, the complete cutting process is decomposed into several processes of removing micro-elements.

[0103] According to the time-domain discretization process of the cutting force signal, the time length of each short time domain is T. i Then, the material removal volume of the discrete infinitesimal element corresponding to the tool trajectory working condition in the short time domain is:

[0104] S i =V·T i

[0105] Where V is the volume of material removed per unit time;

[0106] The machining process is discretized along the tool path signal into N equally timed cutting micro-elements S. i Then the set of cutting infinitesimal elements is:

[0107] S(N)={S i |S1,S2,...,S N}

[0108] Where N also represents the number of spatially discrete cutting micro-elements corresponding to the machining process, and is numerically equal to the number in the short time domain;

[0109] Among them, the operating conditions can be divided into steady operating conditions, time-variable operating conditions, and space-variable operating conditions.

[0110] Steady operating conditions include machine tool conditions, fixture information, and other operating conditions that are considered to remain unchanged over time, denoted as D;

[0111] Time-variable operating conditions include spindle speed, depth of cut, feed rate, tool axis vector, etc., and have spatially variable characteristics, denoted as C. t (p);

[0112] Variable working conditions refer to workpiece geometric and material characteristics that change depending on the machining position of the part, denoted as C. p (t).

[0113] The set of obtained instantaneous machining trajectories is called the cutting micro-element set. In each working condition, the micro-element C... i,j,k Within this, there are machining conditions corresponding to the cutting trajectory, denoted as {D,C}. t (p),C p (t)}.

[0114] Where i, j, and k represent the spatial parameters of the tool path fitting function, and these three values ​​are used to uniquely determine a certain cutting micro-element discrete along the tool path. C 1,1,1

[0115] Step 3: Define the cutting force fluctuation characteristics;

[0116] For each short time domain T i The fluctuation characteristics of the cutting force of all micro-elements are expressed by combining "main features and auxiliary features".

[0117] Step 3.1: Among all the fluctuation characteristics of micro-element cutting force, select several fluctuation characteristics with the best data dispersion, and select one of them as the main feature of micro-element cutting force fluctuation;

[0118] In this embodiment, variance in the fluctuation feature is a measure of the degree of data dispersion. With the same sample size, the larger the variance, the greater the data fluctuation. Therefore, the present invention selects variance as the main feature of cutting force fluctuation.

[0119] The calculation formula is:

[0120]

[0121] Among them, {x i |x1,x2,...,x n} represents a time series signal;

[0122] Step 3.2: Select auxiliary features;

[0123] Step 3.2: Among all the fluctuation characteristics of micro-element cutting force, select the features that have a high correlation with the main features selected in Step 3.1, and then select the features among these features whose correlation with the main features in Step 3.1 reaches the set requirements as auxiliary features of micro-element cutting force fluctuation.

[0124] Define the feature parameters of auxiliary features and their expression formulas, including time-domain features and frequency-domain features;

[0125] Temporal characteristics:

[0126] Time-domain information describes the waveform of a signal with time as the variable. This embodiment extracts 13 types of time-domain features, excluding variance. The signal data is represented as shown in Table 1.

[0127] Table 1. Time-domain characteristic parameters and formulas

[0128]

[0129]

[0130] Frequency domain characteristics:

[0131] Frequency domain analysis is the process of determining the frequency components and signal strength of each frequency band of a monitored signal. This embodiment uses power spectrum information to solve for the signal's power spectrum, employing the following four types of frequency domain features, as shown in Table 2:

[0132] Table 2 Frequency Domain Characteristic Parameters and Formulas

[0133]

[0134] Step 4: Use feature fusion methods to describe cutting force fluctuations and construct a cutting force fluctuation model;

[0135] Step 4.1: Construct the feature evaluation matrix based on the Pearson correlation coefficient method;

[0136] Since a single feature is difficult to comprehensively and accurately describe the characteristics of cutting force fluctuation, and multiple features will cause feature redundancy, this embodiment uses a feature fusion method based on the Pearson correlation coefficient method to describe the characteristics of cutting force fluctuation.

[0137] The Pearson correlation coefficient is used to measure the degree of correlation between two variables X and Y. Its value range is [-1, 1]. The closer the absolute value is to 1, the stronger the correlation between the two variables. The closer the absolute value is to 0, the weaker the correlation between the two variables. The positive or negative value represents the positive or negative relationship of the correlation.

[0138] First, calculate the correlation coefficient between each feature and other features. The calculation formula is as follows:

[0139]

[0140] Then, construct the feature evaluation matrix for all micro-element cutting force fluctuation characteristics;

[0141] This embodiment extracts 18 micro-element cutting force fluctuation features. By calculating the correlation coefficient between each feature and other features, a feature evaluation matrix of the 18 micro-element cutting force fluctuation features is constructed:

[0142]

[0143] Where, r m,n Let be the correlation coefficient between the m-th feature and the n-th feature;

[0144] The feature evaluation matrix heatmap in this embodiment is as follows: Figure 7 As shown.

[0145] Step 4.2: Sort all auxiliary features in descending order of their correlation coefficient with the main feature. Perform feature fusion on the main and auxiliary features. Construct a cutting force fluctuation model based on the weight coefficients of the auxiliary fluctuation features. The calculation formula is as follows:

[0146]

[0147] This example selects a data segment from 0 to 80 seconds. After missing value imputation, standardization, and filtering / denoising, the feature visualization results, sorted from high to low by the correlation coefficient between the 18 features and variance, are shown below. Figure 8 As shown.

[0148] Based on the magnitude of the correlation coefficient with variance, nine features were finally selected as auxiliary fluctuation features: standard deviation, peak-to-peak value, root mean square frequency, waveform factor, frequency variance, centroid frequency, margin factor, impulse factor, and peak factor. Their weighting coefficients are shown in Table 3.

[0149] Table 3. Secondary Fluctuation Characteristic Indicators and Corresponding Weighting Coefficients

[0150]

[0151] After feature fusion, the final cutting force fluctuation model function established in this embodiment is:

[0152]

[0153] The cutting force fluctuation characteristic diagram obtained by selecting data over a certain period of time is shown below. Figure 9 As shown, this illustrates the fluctuations in cutting force over that time period.

[0154] Step 5: Establish a correlation model between cutting force fluctuation information and working condition information;

[0155] Based on the isochronous and isospace properties of the two, a correlation model between cutting force fluctuation information and working condition information is established using time-series matching and coordinate matching methods, so that the two can correspond one-to-one in time and space. According to the established correlation model, the two form a whole, which can both forward index the corresponding machining result from the machining condition and reverse trace the machining condition from the machining result, which is convenient for monitoring the machining process.

[0156] Includes the following sub-steps:

[0157] Step 5.1: Based on the total data acquisition time, the cutting force data is discretized in the time domain according to the order of the machining process to obtain the data set of time-varying elements;

[0158] The duration and number of time-varying elements are matched with the short time domain discretization along the time dimension in step 2.1, that is, the number N of time-varying elements is equal to the short time domain T within the total data acquisition time. i The number of;

[0159] Since the actual collected cutting force data is distributed along the time dimension, the data change scale can be considered very small within an extremely short time period. That is, within this very small time element, the data is uniform and without abrupt changes. Therefore, by using a manually set time element length, all cutting process data can be divided into time dimensions, ultimately obtaining a time element data set of length N.

[0160] Step 5.2: Obtain the cutting tool path according to the machine tool code, and spatially discretize it along the tool path to obtain a set of working condition elements;

[0161] The number of working condition elements is equal to the number of cutting micro-elements discretized along the tool path in step 2.2. The number of cutting micro-elements is consistent with the number of time micro-elements, so that the working condition elements can correspond one-to-one with the time micro-elements.

[0162] Step 5.3: As Figure 10 As shown, discrete cutting force time-domain data contains machining time information, and discrete working condition volumetric model also corresponds to the machining position at a certain moment. Since the two are consistent in quantity and length, and there is a one-to-one correspondence in time sequence, the discrete cutting force time-domain data and discrete working condition volumetric model are aligned in time to achieve the correlation between time and space.

[0163] Taking a planar corner milling process as an example, by discretizing and spatiotemporally correlating cutting force data and working condition data, and then extracting fluctuation characteristics from all short-term cutting force signals, a spatiotemporal distribution model of cutting force fluctuations is established, which facilitates the analysis of cutting force variation trends. For example... Figure 11 As shown in the figure, regions 1, 2, and 3 represent the cutting force fluctuation of the cutting micro-element at different locations and under different working conditions. The plane formed by the x and y axes represents the milled surface of the workpiece, and the z-axis is used to measure the magnitude of the fluctuation value. The red line in the figure represents the tool cutting trajectory, and the colored bar chart distributed along the tool trajectory represents the cutting force fluctuation at the corresponding cutting position. From blue to red, it represents the relative magnitude of the cutting force fluctuation value.

[0164] The fluctuation distribution model shows that the fluctuations are large in the two transition stages between the straight line and the corner in corner milling. This is because the radial depth of cut of the tool changes significantly during this process, resulting in large fluctuations in the cutting force.

[0165] In summary, this method first preprocesses the raw cutting force data monitored during machining, then extracts several time-domain and frequency-domain features. The correlation between these features and variance is calculated using Pearson correlation coefficient analysis, and nine features with high correlation are selected as auxiliary fluctuation features. These features are then fused with the main features to form the cutting force fluctuation features. Subsequently, the fluctuation features are fused with discrete working condition data to construct a cutting force fluctuation model. This improves data visualization in the monitoring of CNC machining processes and provides a foundation for ensuring workpiece quality.

[0166] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the scope of the technology disclosed in the present invention, and such modifications or substitutions should all be covered within the scope of protection of the present invention.

Claims

1. A method for constructing a cutting force fluctuation model based on feature extraction, characterized in that, Includes the following steps: Step 1: Preprocess the raw cutting data; Step 2: Subdivide the cutting force signal, tool path, and operating condition signal; including the following steps: Step 2.1: Subdivide the cutting force signal according to the time-domain discretization concept; Discretize the time domain of a processing procedure into N short time domains T with equal time intervals. i The discretized short-time domain set is as follows: Where N represents the number of short-time domains corresponding to the processing procedure. ; Step 2.2: Subdivide the tool trajectory condition signal according to the spatial discretization concept; The machining process along the tool path signal is discretized into M equally timed cutting micro-elements. Then the set of cutting infinitesimal elements is: Where M represents the number of spatially discrete cutting micro-elements corresponding to the machining process, and is numerically equal to the number in the short time domain; Step 3: Define the cutting force fluctuation characteristics; For each short time domain The fluctuation characteristics of the cutting force of all micro-elements are expressed by combining "main features and auxiliary features", including the following sub-steps: Step 3.1: Among all the fluctuation characteristics of micro-element cutting force, select several fluctuation characteristics with the best data dispersion, and select one of them as the main feature of micro-element cutting force fluctuation; Step 3.2: Among all the fluctuation characteristics of micro-element cutting force, select the features that are highly correlated with the main features selected in Step 3.

1. Then, select the features among these features whose correlation with the main features in Step 3.1 reaches the set requirements as auxiliary features of micro-element cutting force fluctuation, and define the feature parameters and expression formulas of the auxiliary features. The auxiliary features include time-domain features and frequency-domain features; Step 4: Use feature fusion methods to describe cutting force fluctuations and construct a cutting force fluctuation model; this includes the following steps: Step 4.1: Construct the feature evaluation matrix based on the Pearson correlation coefficient method; For all the fluctuation characteristics of the micro-element cutting force, first calculate the correlation coefficient of each characteristic with other characteristics, and then construct the feature evaluation matrix of all the micro-element cutting force fluctuation characteristics; Step 4.2: Sort all auxiliary features from high to low according to their correlation coefficient with the main feature, and perform feature fusion on the main feature and auxiliary features to construct the cutting force fluctuation model; Step 5: Establish a correlation model between cutting force fluctuation information and working condition information using time-series matching and coordinate matching methods; including the following steps: Step 5.1: Set the total data acquisition time, and discretize the cutting force data in the time domain according to the order of the machining process to obtain the data set of time-varying elements; Among them, the time length, number and temporal sequence of the time micro-element are all matched with the short time domain discretization along the time dimension in step 2.1; Step 5.2: Based on the read machine tool code, obtain the cutting tool path, and spatially discretize the cutting force condition along the tool path to obtain a set of condition volume elements; The number of working condition elements is equal to the number of cutting micro-elements discretized along the tool path in step 2.2; Step 5.3: By aligning the discrete cutting force time-domain data and the discrete working condition volumetric model in time, a correlation model between cutting force fluctuation information and working condition information is established.

2. The method for constructing a cutting force fluctuation model based on feature extraction according to claim 1, characterized in that, Step 1 includes the following sub-steps: Step 1.1: Handle missing values ​​in the cutting data; Step 1.2: Standardize the raw cutting data; The conversion formula is: Where: x represents the original data; The processed data; The standard deviation of the original data; Step 1.3: Process the noise value of the cutting force signal.

3. The method for constructing a cutting force fluctuation model based on feature extraction according to claim 1, characterized in that: The main feature in step 3.1 is variance.

4. The method for constructing a cutting force fluctuation model based on feature extraction according to claim 3, characterized in that: The time-domain features in step 3.2 include: Mean, root mean square value, peak value, kurtosis factor, peak factor, waveform factor, impulse factor, standard deviation, root square amplitude, peak-to-peak value, skewness, kurtosis, and margin factor; The frequency domain features in step 3.2 include: Centroid frequency, average frequency, root mean square frequency, and frequency variance.

Images